Data from: Feeding habits of the Middle Triassic pseudosuchian Batrachotomus kupferzellensis from Germany and palaeoecological implications for archosaurs

Data S1: All prepared fossil material from the Kupferzell and Vellberg-Eschenau fossil lagerstätten (southwestern Germany) housed in the SMNS and MHI collections was examined using a magnifying glass, a dissecting microscope, and lighting in different orientations in order to identify bite traces (e.g., Blumenschine et al. 1996; D’Amore & Blumenschine 2009; Drumheller & Brochu 2014; Drumheller et al. 2020). Bite traces were systematically classified according to their morphology (ichnotaxonomy) following Mikuláš et al. (2006), Jacobsen & Bromley (2009) and Pirrone et al. (2014). In addition, equivalences of the identified ichnotaxa with the taphonomic terms coined by Binford (1981) and extensively used in the literature (e.g., Pobiner 2008; D’Amore & Blumenschine 2009; Njau & Gilbert 2016; Drumheller et al. 2020) are provided. Associations between bite traces and their orientation and location on bones were also surveyed Data S2: Tooth crowns of Batrachotomus from Kupferzell (N = 258) and Vellberg-Eschenau (N = 56), all from the UGM, were analysed accounting for: (1) basic morphological classification by measuring (with a digital calliper) the apicobasal height of the crown (or crown height, CH), the mesiodistal depth and the labiolingual width of the crown at its base (or crown base length, CBL, and crown base width, CBW, respectively; for tooth parameters, see: Smith et al. 2005; Hendrickx et al. 2015); (2) identification of the position within the jaw, distinguishing premaxillary from non-premaxillary teeth (also distinguishing teeth from the dentaries and maxillae, when possible). Tables S1-S9 In order to investigate potential statistical differences regarding bite trace patterns on different bitten taxa (Mastodonsaurus vs. Plagiosuchus vs. Nothosaurus vs. Batrachotomus), body regions (skull vs. teeth vs. free vertebrae vs. sacrum vs. ribs vs. pectoral girdle vs. pelvic girdle vs. limbs) and bone regions (proximal end vs. distal end vs. side of the shaft vs. edge of the shaft vs. bone edge vs. bone centre), we converted the data collected in the bite trace database (Mujal et al. 2021, data S1) into three separate datasets, scoring the absence or presence of each bite trace morphotype for all bones within each group. A small fraction of bones was not included in the analyses (Mujal et al. 2021, data S1) because the elements could not be identified due to their fragmentary nature or preservation. These datasets were analysed using a permutational multivariate analysis of variance (perMANOVA). This method estimates the potential overlap between two or more groups by testing the significance of their distribution on the basis of permutation (9999 iterations) and the Euclidean distance as distance measures. In contrast to parametric tests, perMANOVA does not require normal distribution of the data (Anderson 2001; Hammer & Harper 2006). The spatial relationship between groups is expressed by an F-value and a Bonferroni-corrected p-value. Following Wills et al. (1994), the three datasets were further transformed into Euclidean distance matrices and subjected to a principal coordinate analysis (PCOA). This method reduces multivariate data down to a new set of independent variables (principal coordinates) that are linear combinations of the original set with zero covariance (Hammer & Harper 2006), providing a ‘biting space’. Afterwards, we ran a broken-stick method for each PCOA to obtain the number of principal coordinates (PCO) that contain the relevant amount of total variation (DeVita 1979; Jackson 1993). For these PCOs we applied a linear discriminant analysis (LDA), which reduces the number of PCOs to a smaller set of dimensions by maximizing the separation between the given groups using the Mahalanobis distance. This distance measure is estimated from the pooled within-group covariance matrix, resulting in a linear discriminant classifier and an estimated group assignment for each species. These results were cross-validated using Jackknife resampling (Hammer & Harper 2006; Hammer 2020). Based on the confusion matrix (Stehman 1997), we estimated the error of correct identification.